Case Study: What It Costs to Run One Quantum Computer for a Year
Executive Summary
A research group has budget for a quantum computer and asks the obvious question: what does one actually cost? The purchase price is the smallest number in the answer.
This case study builds the total cost of ownership for a 127-qubit superconducting system over its first year — capital, facilities, consumables, staff, and downtime — and then compares it against cloud access, which is what nearly every group should choose and for reasons that are worth making explicit.
Skills applied
- Building a total cost of ownership model for cryogenic hardware (§29.14).
- Accounting for facilities, consumables, and staff.
- Quantifying availability and its effect on effective cost.
- Comparing ownership against cloud access on the right basis.
Phase 1: Capital
| Item | Cost |
|---|---|
| Dilution refrigerator (with compressors, gas-handling) | $700k |
| Quantum processor chip and package | $150k |
| Control electronics (AWGs, digitizers, ~127 channels) | $900k |
| Cryogenic components (attenuators, amplifiers, cabling) | $250k |
| Racks, wiring, vibration isolation, RF shielding | $120k |
| Classical control server and storage | $60k |
| Total capital | $2.18M |
The control electronics rival the refrigerator, and both exceed the processor by a wide margin. The quantum chip is roughly 7% of the capital cost — an inversion of the intuition that the qubits are the expensive part.
Phase 2: Facilities
Requirements that are easy to overlook until the equipment arrives:
| Requirement | Detail | Cost |
|---|---|---|
| Floor space | ~40 m², reinforced (system weighs ~1,500 kg) | $30k/yr |
| Electrical | 3-phase, ~25 kW continuous | included below |
| Cooling | ~20 kW heat rejection from compressors | $25k build-out |
| Vibration isolation | Pulse tubes and building vibration degrade coherence | $40k one-time |
| RF shielding | Screened room to keep stray microwaves out | $80k one-time |
| Helium recovery | Optional; ³He is expensive and scarce | $60k one-time |
One-time facilities: ~$205k. Recurring: ~$55k/yr.
Vibration isolation is regularly underestimated. A pulse-tube cooler vibrates at ~1.4 Hz, and that vibration modulates qubit frequencies. Groups have installed systems only to discover that a nearby lift or HVAC unit degrades coherence measurably.
Phase 3: Recurring costs
| Item | Annual cost |
|---|---|
| Electricity (25 kW × 8,760 h × $0.12/kWh) | $26k | |
| Helium (³He/⁴He top-ups, losses) | $35k |
| Liquid nitrogen (pre-cooling) | $8k |
| Service contract (refrigerator + electronics) | $180k |
| Consumables and spares | $25k |
| Facilities (from Phase 2) | $55k |
| Subtotal | $329k/yr |
Staff — the largest line item:
| Role | FTE | Annual cost |
|---|---|---|
| Cryogenic engineer | 0.5 | $75k |
| Control/calibration scientist | 1.0 | $160k |
| Software/systems engineer | 1.0 | $150k |
| Research scientist (uses the machine) | 1.0 | $170k |
| Subtotal | 3.5 FTE | $555k/yr |
Total annual operating cost: ~$884k.
Phase 4: Availability — the number that changes everything
A quantum computer is not available 24/7.
| Activity | Time |
|---|---|
| Cooldown from room temperature | 36–72 hours |
| Full calibration after cooldown | 8–24 hours |
| Daily recalibration | 1–3 hours |
| Unplanned downtime (TLS drift, component failure) | ~10% |
| Planned maintenance / warm-up cycles | ~4 weeks/yr |
Realistic first-year availability: 60–70%, and lower in year one while staff learn the system.
At 65% availability, the machine delivers ~5,700 usable hours. Amortizing capital over 5 years:
$$\frac{\$2.18M/5 + \$884k}{5{,}700\ \text{h}} = \frac{\$1.32M}{5{,}700} \approx \mathbf{\$232\ \text{per usable hour}}$$
And that hour is device time, not exclusive research time — shared among the group.
Phase 5: The cloud comparison
| Own the machine | Cloud access | |
|---|---|---|
| Year-1 cost | ~$3.1M (capital + operating) | $50k–$500k depending on usage | |
| Availability | 65%, one device | Multiple devices, multiple vendors |
| Hardware refresh | Buy a new one | Automatic |
| Staff needed | 3.5 FTE | 0 |
| Calibration burden | Yours | Vendor's |
| Queue wait | None | Minutes to hours |
| Access to latest hardware | No | Yes |
| Control over low-level pulses | Full | Limited (though improving) |
For nearly every group, cloud access is the correct choice. The cost difference is roughly an order of magnitude, and cloud users get access to several architectures rather than one.
Ownership is justified when you need pulse-level control for hardware research; you are developing hardware, control electronics, or calibration methods; you need guaranteed uninterrupted access for time-critical experiments; or you have security constraints prohibiting external compute.
Ownership is not justified for algorithm development, applications research, education, or benchmarking — all of which cloud access serves better and cheaper.
Phase 6: The wider point
This exercise makes concrete something the field's marketing obscures. A quantum computer is:
- A cryogenic instrument first. Refrigeration, wiring, and shielding dominate cost and complexity.
- A control-electronics problem second. The electronics cost more than the chip by a factor of six.
- A staffing problem third. People are the largest recurring expense, and expertise is scarce.
- A quantum processor a distant fourth. 7% of capital.
That ordering explains a good deal about the industry: why vendors sell cloud access rather than machines, why hardware companies employ more classical engineers than physicists, and why "how many qubits" is a poor proxy for progress — the difficulty is not principally in the qubits.
Discussion Questions
- The processor is 7% of capital. What does that imply about the "qubit count" framing of progress?
- Availability of 65% raised effective hourly cost substantially. What would you do to improve it, and what would that cost?
- Cloud access is ~10× cheaper for most users. Why do organizations still buy machines?
- Staff is the largest recurring line. What does that suggest about how fast the industry can scale?
Your Turn: Extensions
- Build the same model for a trapped-ion system; note which line items change most.
- Compute the break-even usage level at which ownership beats cloud pricing.
- Estimate the cost per shot for both models and compare against a variational algorithm's budget.
- Research current cloud pricing and recompute Phase 5 with real numbers.
Key Takeaways
- Capital is ~$2.2M, of which the quantum processor is about 7%; control electronics and refrigeration dominate.
- Facilities requirements — vibration isolation, RF shielding, 20 kW of heat rejection — are frequently underestimated.
- Staff at ~3.5 FTE is the largest recurring cost, exceeding all consumables and service contracts combined.
- Realistic availability of 60–70% puts effective cost near $230 per usable device-hour.
- Cloud access is roughly an order of magnitude cheaper and correct for almost everyone except hardware researchers and those needing pulse-level control.